When AI Agents Start Talking to Each Other, Who's in Control?
Your future AI assistant will do business with other AIs — here's the risk.
A new report from Australia’s AI Safety Institute presents a framework for understanding the risks that emerge when AI agents interact with each other—especially as those interactions cross organizational boundaries. It explains that once agents operate beyond an organization’s perimeter, no single organization can fully see, control, or govern them. The report introduces three deployment tiers based on the minimum common governance binding interacting agents: singular governance, federated governance, and open environments. It examines risk factors, failure modes, and available controls, and identifies who can apply them—or where collective action is needed to close gaps.
- AI agents will soon talk to each other outside any single company's control — like your bank's AI negotiating with a supplier's AI.
- The report defines three levels of oversight: one company in charge, several companies sharing rules, or open internet with no authority.
- Crossing organizational lines means no single player can monitor everything, so regulators and industries must act together to prevent AI mishaps.
Why It Matters
As AI agents handle more daily tasks, knowing who's accountable when they interact keeps our money, data, and decisions safe.